Tian Dong (Shanghai Jiao Tong University), Minhui Xue (CSIRO's Data61), Guoxing Chen (Shanghai Jiao Tong University), Rayne Holland (CSIRO's Data61), Yan Meng (Shanghai Jiao Tong University), Shaofeng Li (Southeast University), Zhen Liu (Shanghai Jiao Tong University), Haojin Zhu (Shanghai Jiao Tong University)

Open-source Large Language Models (LLMs) have recently gained popularity because of their comparable performance to proprietary LLMs. To efficiently fulfill domain-specialized tasks, open-source LLMs can be refined, without expensive accelerators, using low-rank adapters. However, it is still unknown whether low-rank adapters can be exploited to control LLMs. To address this gap, we demonstrate that an infected adapter can induce, on specific triggers, an LLM to output content defined by an adversary and to even maliciously use tools. To train a Trojan adapter, we propose two novel attacks, POLISHED and FUSION, that improve over prior approaches. POLISHED uses a superior LLM to align naïvely poisoned data based on our insight that it can better inject poisoning knowledge during training. In contrast, FUSION leverages a novel over-poisoning procedure to transform a benign adapter into a malicious one by magnifying the attention between trigger and target in model weights. In our experiments, we first conduct two case studies to demonstrate that a compromised LLM agent can use malware to control the system (e.g., a LLM-driven robot) or to launch a spear-phishing attack. Then, in terms of targeted misinformation, we show that our attacks provide higher attack effectiveness than the existing baseline and, for the purpose of attracting downloads, preserve or improve the adapter’s utility. Finally, we designed and evaluated three potential defenses. However, none proved entirely effective in safeguarding against our attacks, highlighting the need for more robust defenses supporting a secure LLM supply chain.

View More Papers

Oreo: Protecting ASLR Against Microarchitectural Attacks

Shixin Song (Massachusetts Institute of Technology), Joseph Zhang (Massachusetts Institute of Technology), Mengjia Yan (Massachusetts Institute of Technology)

Read More

Magmaw: Modality-Agnostic Adversarial Attacks on Machine Learning-Based Wireless Communication...

Jung-Woo Chang (University of California, San Diego), Ke Sun (University of California, San Diego), Nasimeh Heydaribeni (University of California, San Diego), Seira Hidano (KDDI Research, Inc.), Xinyu Zhang (University of California, San Diego), Farinaz Koushanfar (University of California, San Diego)

Read More

Siniel: Distributed Privacy-Preserving zkSNARK

Yunbo Yang (The State Key Laboratory of Blockchain and Data Security, Zhejiang University), Yuejia Cheng (Shanghai DeCareer Consulting Co., Ltd), Kailun Wang (Beijing Jiaotong University), Xiaoguo Li (College of Computer Science, Chongqing University), Jianfei Sun (School of Computing and Information Systems, Singapore Management University), Jiachen Shen (Shanghai Key Laboratory of Trustworthy Computing, East China Normal…

Read More